A method for intelligent processing of multi-source sensor data based on cylinder bracket
By performing multi-source intelligent processing on cylinder bracket sensor data and adaptively correcting it by combining abnormal prominence indicators and the prominence of transmission relationships, the problem of misjudgment of noise data is solved, and the accuracy of cylinder anomaly detection is improved.
Patent Information
- Application Number
- CN202410682408.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-05-29
- Publication Date
- 2024-11-26
- Estimated Expiration
- 2044-05-29
AI Technical Summary
In existing technologies, when anomaly detection algorithms are used to detect anomalies in cylinder bracket sensor data, noise data is easily misjudged as cylinder anomalies, resulting in poor detection accuracy.
A multi-source sensor data intelligent processing method based on cylinder bracket is adopted. By obtaining the initial anomaly score of the sensor data, and combining the anomaly prominence index, the coordination coefficient and the prominence of the transmission relationship for adaptive correction, the target anomaly score is obtained. Taking into account the stability and correlation of the sensor data, the detection accuracy is improved.
This effectively avoids misjudgment of noise data, improves the accuracy of sensor data and cylinder anomaly detection, and ensures that the detection results more accurately reflect the true abnormal situation of the cylinder.
Smart Images

Figure CN118690290B_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of sensor data processing, and in particular to a multi-source sensor data intelligent processing method based on a cylinder bracket. Background Art
[0002] With the development of science and technology, the application of sensor data processing is becoming more and more extensive. For example, the sensor data based on the cylinder bracket can be processed by abnormality detection. If the sensor data is abnormal, it is determined that the cylinder installed on the cylinder bracket is abnormal. At present, when performing abnormality detection on data, the method usually used is: calculate the abnormality score of the data through the abnormality detection algorithm, and judge whether the data is abnormal based on the abnormality score of the data. Among them, the abnormality score of the data often represents the degree of outlier of the data.
[0003] However, when using anomaly detection algorithms to detect anomalies in sensor data collected from cylinders mounted on cylinder brackets, the following technical problems often occur:
[0004] When calculating the anomaly score of sensor data through anomaly detection algorithms, only the density distribution of sensor data is often considered. The more outliers the density distribution of sensor data is, the more likely it is to be abnormal data, and thus the cylinder installed on the cylinder bracket is more likely to have an abnormality. However, due to the influence of environmental factors, some sensors may have poor contact at certain times, resulting in noise data in the collected sensor data, and noise data often also appears to be outliers. Therefore, when anomaly detection is performed directly on sensor data through anomaly detection algorithms, noise data may be misjudged as cylinder abnormal data, resulting in poor accuracy in anomaly detection of sensor data, and further poor accuracy in cylinder abnormality detection. Summary of the invention
[0005] In order to solve the technical problem of poor accuracy in cylinder abnormality detection due to poor accuracy in abnormality detection of sensor data, the present invention proposes an intelligent processing method for multi-source sensor data based on a cylinder bracket.
[0006] The present invention provides a multi-source sensor data intelligent processing method based on a cylinder bracket, the method comprising:
[0007] Obtain sensor data of each preset dimension collected by each cylinder to be detected installed on the target cylinder bracket within the current time period, and determine the initial anomaly score corresponding to each sensor data through an anomaly detection algorithm;
[0008] According to the abnormal prominent index of each cylinder to be detected in each two preset dimensions and all sensor data of each cylinder to be detected in each two preset dimensions collected in the current time period, the current coordination coefficient between each cylinder to be detected and each two preset dimensions is determined, wherein the abnormal prominent index is characterized by the stability of the sensor data in the preset dimensions;
[0009] Determine the prominence of the conduction relationship between each two cylinders to be tested according to the abnormal prominence index of each two cylinders to be tested in all preset dimensions and the angle between the total force directions corresponding to each two cylinders to be tested;
[0010] According to the current synergy coefficient between the cylinders to be detected in the preset dimensions and the prominence of the conduction relationship between the cylinders to be detected, the initial anomaly score corresponding to each sensor data is adaptively corrected to obtain the target anomaly score corresponding to each sensor data;
[0011] Anomaly detection is performed on all sensor data according to the target anomaly scores corresponding to all sensor data.
[0012] Optionally, a method for obtaining an abnormal prominent index of each cylinder to be detected in each preset dimension includes:
[0013] According to all sensor data in each preset dimension collected by each cylinder to be detected in the current time period, an abnormal prominent index of each cylinder to be detected in each preset dimension is determined.
[0014] Optionally, the determining of the abnormal prominent index of each cylinder to be detected in each preset dimension according to all sensor data in each preset dimension collected by each cylinder to be detected in the current time period includes:
[0015] The absolute value of the difference between each adjacent two sensor data in each preset dimension collected by each cylinder to be detected in the current time period is determined as the target data difference, so as to form a target data difference set of each cylinder to be detected in each preset dimension;
[0016] The mean of all target data differences in the target data difference set of each cylinder to be detected under each preset dimension is determined as the data representative difference of each cylinder to be detected under each preset dimension, wherein the data representative difference represents the average fluctuation degree of the sensor data under the preset dimension;
[0017] The information entropy of all sensor data in each preset dimension collected by each cylinder to be detected in the current time period is determined as a data confusion factor in each preset dimension of each cylinder to be detected, wherein the data confusion factor represents the confusion of the sensor data in the preset dimension;
[0018] Determine the range of all sensor data in each preset dimension collected by each cylinder to be detected in the current time period as the target range of each cylinder to be detected in each preset dimension, wherein the target range represents the maximum fluctuation degree of the sensor data in the preset dimension;
[0019] According to the data representative difference, data confusion factor and target range of each cylinder to be tested in each preset dimension, the abnormal prominence index of each cylinder to be tested in each preset dimension is determined, wherein the data representative difference, data confusion factor and target range are all negatively correlated with the abnormal prominence index.
[0020] Optionally, the determining of the current coordination coefficient between each cylinder to be detected and each two preset dimensions according to the abnormal prominent index of each cylinder to be detected and each two preset dimensions collected by the cylinder to be detected in the current time period includes:
[0021] Determine any one of the cylinders to be detected as a marked cylinder, and respectively determine any two preset dimensions as a first marked dimension and a second marked dimension;
[0022] Taking a time series consisting of all sensor data under a first marking dimension collected by the marked cylinder in a current time period as a first data series;
[0023] Taking a time series consisting of all sensor data in a second marking dimension collected by the marked cylinder in a current time period as a second data series;
[0024] determining the absolute value of the Pearson correlation coefficient between the first data sequence and the second data sequence as a dimensional similarity factor between the first marking dimension and the second marking dimension of the marked cylinder;
[0025] Based on the dimensional similarity factor between the marked cylinder in the first marked dimension and the second marked dimension, and the abnormal prominence index of the marked cylinder in the first marked dimension and the second marked dimension, the current synergy coefficient between the marked cylinder in the first marked dimension and the second marked dimension is determined, wherein the abnormal prominence index and the dimensional similarity factor are both positively correlated with the current synergy coefficient.
[0026] Optionally, determining the prominence of the conduction relationship between each two cylinders to be detected according to the abnormal prominence index of each two cylinders to be detected in all preset dimensions and the angle between the total force directions corresponding to each two cylinders to be detected includes:
[0027] The acquisition time corresponding to the sensor data is used as the horizontal coordinate, and the sensor data is used as the vertical coordinate to construct the coordinate point corresponding to the sensor data;
[0028] Performing curve fitting on coordinate points corresponding to all sensor data in each preset dimension collected by each cylinder to be detected in the current time period to obtain a data fitting curve of each cylinder to be detected in each preset dimension;
[0029] The slope of the coordinate point corresponding to each sensor data on the data fitting curve to which it belongs is determined as the target slope corresponding to each sensor data, wherein the target slope represents the fluctuation change of the sensor data;
[0030] The prominence of the conduction relationship between each two cylinders to be detected is determined based on the abnormal prominence indicators of each two cylinders to be detected in all preset dimensions, the angle between the total force directions corresponding to each two cylinders to be detected, and the target slope corresponding to all sensor data collected by each two cylinders to be detected in the current time period.
[0031] Optionally, the formula corresponding to the prominence of the conduction relationship between any two cylinders to be detected is:
[0032]
[0033] Among them, G ix is the salience of the conduction relationship between the i-th cylinder to be tested and the x-th cylinder to be tested; i and x are the serial numbers of different cylinders to be tested; norm() is the normalization function; n is the number of preset dimensions; j is the serial number of the preset dimension; β is a pre-set factor greater than 0; H ij is the abnormal outstanding index of the i-th cylinder to be tested under the j-th preset dimension; H xj is the abnormal prominence index of the xth cylinder to be tested under the jth preset dimension; Δk ixj Characterizes the difference in data change between the i-th cylinder to be tested and the x-th cylinder to be tested in the j-th preset dimension; || is the absolute value function; θ ix is the angle between the total force direction corresponding to the i-th cylinder to be tested and the total force direction corresponding to the x-th cylinder to be tested; π is the radian value of 180°; M is the number of acquisition moments in the current time period; b is the sequence number of the acquisition moment in the current time period; k ijb is the target slope corresponding to the sensor data of the i-th cylinder to be tested under the j-th preset dimension collected at the b-th collection time in the current time period; k xjb It is the target slope corresponding to the sensor data of the x-th cylinder to be detected under the j-th preset dimension collected at the b-th collection time in the current time period.
[0034] Optionally, the adaptive correction of the initial abnormality score corresponding to each sensor data according to the current synergy coefficient between the cylinders to be detected in the preset dimensions and the prominence of the conduction relationship between the cylinders to be detected to obtain the target abnormality score corresponding to each sensor data includes:
[0035] Determine any sensor data as label data, determine the cylinder to be detected to which the label data belongs as a reference cylinder, and determine the preset dimension to which the label data belongs as a reference dimension;
[0036] The target abnormality score corresponding to the labeled data is determined based on the initial abnormality score corresponding to the labeled data, the prominence of the conduction relationship between the reference cylinder and other cylinders to be detected, and the current coordination coefficient between the reference cylinder and other preset dimensions.
[0037] Optionally, the formula corresponding to the target anomaly score corresponding to the sensor data is:
[0038] E ijb =norm(S ijb ×(w ijb1 +w ijb2 ));
[0039]
[0040] Among them, E ijb is the target anomaly score corresponding to the sensor data of the i-th cylinder to be detected under the j-th preset dimension collected at the b-th collection time in the current time period. The preset dimension to which the sensor data belongs is the j-th preset dimension, and the cylinder to be detected to which the sensor data belongs is the i-th cylinder to be detected; i and x are the serial numbers of different cylinders to be detected; b is the serial number of the collection time in the current time period; j and r are the serial numbers of different preset dimensions; norm() is the normalization function; S ijb is the initial abnormal score corresponding to the sensor data of the i-th cylinder to be tested under the j-th preset dimension collected at the b-th collection time in the current time period; w ijb1 and w ijb2 It is used to correct S ijb The two weights of; N is the number of cylinders to be tested; S xjb is the initial abnormality score corresponding to the sensor data of the xth cylinder to be tested under the jth preset dimension collected at the bth collection time in the current time period; G ix is the salience of the conduction relationship between the i-th cylinder to be tested and the x-th cylinder to be tested; n is the number of preset dimensions; S irbis the initial abnormal score corresponding to the sensor data of the i-th cylinder to be detected under the r-th preset dimension collected at the b-th collection time in the current time period; μ ijr is the current coordination coefficient between the jth preset dimension and the rth preset dimension of the i-th cylinder to be detected.
[0041] Optionally, performing anomaly detection on all sensor data according to target anomaly scores corresponding to all sensor data includes:
[0042] If the target anomaly score corresponding to the sensor data is greater than the preset anomaly threshold, the sensor data is determined to be abnormal data.
[0043] Optionally, the method further comprises:
[0044] If abnormal data exists in the abnormal detection result of abnormal detection of all sensor data, it is determined that there is an abnormal cylinder among the cylinders to be detected mounted on the target cylinder bracket.
[0045] The present invention has the following beneficial effects:
[0046] The present invention provides a multi-source sensor data intelligent processing method based on a cylinder bracket, which realizes abnormal detection of multi-source sensor data by processing multi-source sensor data, solves the technical problem of poor accuracy of cylinder abnormal detection caused by poor accuracy of abnormal detection of sensor data, and improves the accuracy of abnormal detection of multi-source sensor data and cylinder. Among them, sensor data under multiple preset dimensions can represent multi-source sensor data. In general, there is often a certain correlation between different cylinders on the same cylinder bracket. For example, if a cylinder on the cylinder bracket fails, it may cause uneven force on the cylinder bracket, which may cause abnormalities in other cylinders. In other words, when a cylinder on the cylinder bracket is abnormal, the sensor data representing the real situation of the cylinder often has abnormalities, and the sensor data representing the real situation of other cylinders that are related to the cylinder often also has certain abnormalities. Therefore, the sensor data representing the real abnormality of the cylinder abnormality often does not exist in isolation, and when an abnormality due to misjudgment of noise data occurs, other sensor data often does not also have abnormalities. Therefore, compared with directly performing anomaly detection on sensor data through anomaly detection algorithm, the present invention comprehensively considers the relationship between preset dimensions and the relationship between cylinders to be detected, such as the current synergy coefficient and the prominence of the conduction relationship, and adaptively corrects the initial anomaly score corresponding to each sensor data. This can avoid misjudging noise data as sensor data of cylinder abnormality to a certain extent, thereby improving the accuracy of anomaly detection on sensor data, and further improving the accuracy of anomaly detection on cylinders to be detected. BRIEF DESCRIPTION OF THE DRAWINGS
[0047] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0048] Figure 1 It is a flow chart of a multi-source sensor data intelligent processing method based on a cylinder bracket of the present invention;
[0049] Figure 2 is a flow chart of a method for determining a current synergy coefficient of the present invention;
[0050] Figure 3 A flow chart of a method for determining an abnormal prominence index of the present invention;
[0051] Figure 4 The flowchart of the method for determining the prominence of the conductive relationship of the present invention. DETAILED DESCRIPTION
[0052] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the specific implementation methods, structures, features and effects of the technical solutions proposed by the present invention are described in detail below in conjunction with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures or characteristics in one or more embodiments may be combined in any suitable form.
[0053] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.
[0054] refer to Figure 1 , shows the process of some embodiments of a method for intelligently processing multi-source sensor data based on a cylinder bracket according to the present invention. The method for intelligently processing multi-source sensor data based on a cylinder bracket comprises the following steps:
[0055] Step S1, obtaining sensor data of each preset dimension collected by each cylinder to be detected installed on the target cylinder bracket within the current time period, and determining the initial anomaly score corresponding to each sensor data through an anomaly detection algorithm.
[0056] Among them, the target cylinder bracket may be a cylinder bracket on which a cylinder is installed. The cylinder bracket may be a structural component for supporting and fixing the cylinder. The cylinder to be detected may be a cylinder installed on the target cylinder bracket. The current time period may be a preset time period with the current moment as the end time and a relatively short time period. For example, the duration corresponding to the current time period may be one day. The preset dimension may be a preset dimension related to the abnormality of the cylinder. For example, the preset dimension may be but is not limited to: temperature dimension, vibration dimension, force dimension and actual load dimension. The sensor data may be a preset dimension value collected by the sensor. For example, if the preset dimension is a temperature dimension, the sensor data under the preset dimension may be a temperature value. The anomaly detection algorithm may be an algorithm for performing data anomaly detection. For example, the anomaly detection algorithm may be an HBOS (Histogram-Based Outlier Score, linear complexity anomaly detection) algorithm. The initial anomaly score may be an anomaly score obtained by performing anomaly detection on the sensor data through the anomaly detection algorithm.
[0057] It should be noted that quantifying the initial anomaly score corresponding to each sensor data can facilitate the subsequent comprehensive consideration of the relationship between different preset dimensions and the relationship between the cylinders to be detected, so as to correct the initial anomaly score corresponding to each sensor data, thereby improving the final anomaly detection effect. Secondly, the cylinder bracket of the embodiment of the present invention can be used for the Ilizarov technique. For example, part of the screw used in the Ilizarov technique can be replaced with a cylinder.
[0058] As an example, taking the temperature dimension as an example, a cylinder to be detected is recorded as a candidate cylinder. The method for collecting sensor data under the temperature dimension of the candidate cylinder in the current time period may be: in the current time period, the temperature value of the candidate cylinder is collected once every two seconds through the temperature sensor as the sensor data under the temperature dimension. At this time, the sensor data under the temperature dimension collected by the candidate cylinder in the current time period may include: the temperature value of the candidate cylinder at different times collected in the current time period. Among them, the start time of the current time period may be the collection time of the first collection of sensor data.
[0059] It should be noted that, since the method for collecting sensor data under each preset dimension in the current time period for each cylinder to be detected is the same, the embodiment of the present invention lists a method for collecting sensor data under a preset dimension, and the methods for collecting sensor data under other preset dimensions can refer to the listed collection method.
[0060] Step S2, determining the current coordination coefficient between each cylinder to be detected and each two preset dimensions according to the abnormal prominent index of each cylinder to be detected and each two preset dimensions collected by the sensor data of each cylinder to be detected in the current time period.
[0061] Among them, the abnormal prominence index can be characterized by the stability of sensor data under a preset dimension.
[0062] As an example, the method for determining the current synergy coefficient between each two preset dimensions of each cylinder to be detected can be as follows: Figure 2 As shown, the following steps may be specifically included:
[0063] Step 201, based on all sensor data of each cylinder to be detected under each preset dimension collected in the current time period, a method for determining the abnormal prominent index of each cylinder to be detected under each preset dimension can be as follows: Figure 3 As shown, the following steps may be specifically included:
[0064] Step 301 , determining the absolute value of the difference between each adjacent two sensor data in each preset dimension collected by each cylinder to be detected in the current time period as the target data difference, and forming a target data difference set of each cylinder to be detected in each preset dimension.
[0065] The two adjacent sensor data may be two sensor data that are adjacent at the time of collection.
[0066] For example, taking the temperature dimension as an example, a cylinder to be detected is recorded as a candidate cylinder. If the temperature values of the candidate cylinder at three moments are collected in the current time period and are recorded as the first temperature value, the second temperature value and the third temperature value respectively, the absolute value of the difference between the first temperature value and the second temperature value can be used as the first target data difference; the absolute value of the difference between the second temperature value and the third temperature value can be used as the second target data difference; at this time, the target data difference set of the candidate cylinder in the temperature dimension may include: the first target data difference and the second target data difference.
[0067] Step 302 : Determine the mean of all target data differences in the target data difference set of each cylinder to be detected in each preset dimension as the data representative difference of each cylinder to be detected in each preset dimension.
[0068] Among them, the data representative difference can characterize the average fluctuation degree of sensor data under the preset dimension.
[0069] Step 303 , determining the information entropy of all sensor data in each preset dimension collected by each cylinder to be detected in the current time period as the data confusion factor of each cylinder to be detected in each preset dimension.
[0070] Among them, the data confusion factor can characterize the confusion of sensor data under a preset dimension.
[0071] Step 304 , determining the range of all sensor data in each preset dimension collected by each cylinder to be detected in the current time period as the target range of each cylinder to be detected in each preset dimension.
[0072] Among them, the target range can represent the maximum fluctuation degree of sensor data under the preset dimension.
[0073] Step 305 , determining the abnormal prominent index of each cylinder to be detected in each preset dimension according to the data representative difference, data confusion factor and target range of each cylinder to be detected in each preset dimension.
[0074] Among them, data representative difference, data confusion factor and target range can all be negatively correlated with abnormal prominence indicators.
[0075] For example, the formula for determining the abnormal prominence index of the cylinder to be tested under the preset dimension may be:
[0076] Among them, H ij is the abnormal outstanding index of the i-th cylinder to be tested under the j-th preset dimension. i is the serial number of the cylinder to be tested. j is the serial number of the preset dimension. exp() is an exponential function with a natural constant as the base. A ij is the data representative difference of the i-th cylinder to be tested under the j-th preset dimension. N ij is the number of different sensor data under the jth preset dimension collected by the i-th cylinder to be tested in the current time period. a is the number of different sensor data under the jth preset dimension collected by the i-th cylinder to be tested in the current time period. n ij is the total number of all sensor data collected by the i-th cylinder to be tested under the j-th preset dimension in the current time period. ija It is the number of a-th different sensor data under the j-th preset dimension collected by the i-th cylinder to be detected in the current time period. Based on the natural constants The logarithm of . is the data confusion factor of the i-th cylinder to be tested in the j-th preset dimension. ΔT ij is the target range of the i-th cylinder to be tested in the j-th preset dimension.
[0077] It should be noted that, in general, it is often necessary to perform real-time abnormality detection on the cylinder to be detected. Therefore, in the process of abnormality detection of the cylinder, the cylinder in a short period of time before the current moment is often operating normally. If the cylinder has already failed or abnormal in the short period of time before the current moment, the abnormality detection process is often stopped at the current stage and the staff is reminded that the cylinder needs to be repaired. Therefore, the cylinder to be detected in the time period before the current moment is often operating normally. Therefore, if the fluctuation of the sensor data under a preset dimension in the current time period is small, it often means that the fluctuation of the data under the preset dimension is small when the cylinder to be detected is normal. At this time, if the outlier degree of a certain sensor data under the preset dimension is large, it often means that the sensor data is more likely to be abnormal data, and the degree of abnormal prominence under the preset dimension is greater. It often means that the abnormality score of the sensor data under the preset dimension obtained by the abnormality detection algorithm can better represent the actual situation of the cylinder abnormality. When A ij The larger the value, the greater the data fluctuation under the jth preset dimension when the i-th cylinder to be tested is normal. When ΔT is larger, it often means that when the i-th cylinder to be tested is normal, the data confusion degree in the j-th preset dimension is greater. ij The larger the value, the greater the data change degree under the jth preset dimension when the i-th cylinder to be tested is normal. ij When the value is larger, it often means that when the i-th cylinder to be tested is normal, the data fluctuation under the j-th preset dimension is small, which often means that the anomaly score of the sensor data under the j-th preset dimension obtained by the anomaly detection algorithm can better represent the actual situation of the cylinder abnormality.
[0078] Step 202: determine any one of the cylinders to be detected as a marked cylinder, and determine any two preset dimensions as a first marked dimension and a second marked dimension, respectively.
[0079] Step 203: Use the time series consisting of all sensor data under the first marking dimension collected by the marked cylinder in the current time period as the first data series.
[0080] Step 204 , taking a time series consisting of all sensor data under a second marking dimension collected by the marked cylinder in the current time period as a second data series.
[0081] Step 205 , determining the absolute value of the Pearson correlation coefficient between the first data sequence and the second data sequence as the dimensional similarity factor between the marked cylinders in the first marked dimension and the second marked dimension.
[0082] Step 206, determining the current coordination coefficient between the marked cylinders in the first marked dimension and the second marked dimension based on the dimensional similarity factor between the marked cylinders in the first marked dimension and the second marked dimension and the abnormal prominence index of the marked cylinders in the first marked dimension and the second marked dimension.
[0083] Among them, both the abnormal prominence index and the dimension similarity factor can be positively correlated with the current synergy coefficient.
[0084] For example, the formula for determining the current coordination coefficient between any two preset dimensions of the cylinder to be detected may be:
[0085] Among them, μ ijr is the current synergy coefficient between the jth preset dimension and the rth preset dimension of the i-th cylinder to be tested. i is the serial number of the cylinder to be tested. j and r are the serial numbers of different preset dimensions. H ij H is the abnormal prominence index of the i-th cylinder to be tested under the j-th preset dimension. ir is the abnormal prominence index of the i-th cylinder to be tested under the r-th preset dimension. || is the absolute value function. PCCs(t ij ,t ir ) indicates t ij and t ir Pearson correlation coefficient between t ij Represents the time series consisting of all sensor data under the jth preset dimension collected by the i-th cylinder to be tested in the current time period. ir represents the time series composed of all sensor data under the rth preset dimension collected by the i-th cylinder to be tested in the current time period. |PCCs(t ij ,t ir )| is the dimensional similarity factor between the j-th preset dimension and the r-th preset dimension of the i-th cylinder to be detected.
[0086] It should be noted that when |PCCs(t ij ,t ir )| is larger, it often indicates that the data distribution of the i-th cylinder to be tested between the r-th preset dimension and the j-th preset dimension is more similar, and it often indicates that the correlation between the r-th preset dimension and the j-th preset dimension is greater. ij When H is larger, it often means that the abnormal score of the sensor data under the jth preset dimension obtained by the abnormality detection algorithm can better represent the actual situation of cylinder abnormality. ir The larger the value, the more likely it is that the abnormal score of the sensor data under the rth preset dimension obtained by the abnormality detection algorithm can better represent the actual situation of the cylinder abnormality. ijrThe larger it is, the greater the correlation between the r-th preset dimension and the j-th preset dimension is, and the anomaly scores of the sensor data under the r-th preset dimension and the j-th preset dimension obtained by the anomaly detection algorithm can often better represent the actual situation of the cylinder abnormality.
[0087] Step S3, determining the prominence of the conduction relationship between each two cylinders to be detected based on the abnormal prominence index of each two cylinders to be detected in all preset dimensions and the angle between the total force directions corresponding to each two cylinders to be detected.
[0088] The total force direction corresponding to the cylinder to be tested may be the direction of the resultant force corresponding to the cylinder to be tested. The resultant force corresponding to the cylinder to be tested may be the resultant force formed by all the forces acting on the cylinder to be tested. The angle between the total force directions corresponding to the two cylinders to be tested is the minimum positive angle formed by the intersection of the resultant forces corresponding to the two cylinders to be tested. That is, the value range of the angle between the total force directions corresponding to the two cylinders to be tested may be [0, π]. π is a radian value of 180°.
[0089] As an example, the method for determining the prominence of the conduction relationship between each two cylinders to be detected can be as follows: Figure 4 As shown, the following steps may be specifically included:
[0090] Step 401 , using the collection time corresponding to the sensor data as the horizontal coordinate and the sensor data as the vertical coordinate, constructing the coordinate point corresponding to the sensor data.
[0091] Step 402 , curve fitting is performed on the coordinate points corresponding to all sensor data in each preset dimension collected by each cylinder to be detected in the current time period to obtain a data fitting curve of each cylinder to be detected in each preset dimension.
[0092] Step 403 : Determine the slope of the coordinate point corresponding to each sensor data on the data fitting curve to which it belongs as the target slope corresponding to each sensor data.
[0093] Among them, the target slope can represent the fluctuation of sensor data.
[0094] Step 404, determines the prominence of the conduction relationship between each two cylinders to be detected based on the abnormal prominence index of each two cylinders to be detected in all preset dimensions, the angle between the total force directions corresponding to each two cylinders to be detected, and the target slope corresponding to all sensor data collected by each two cylinders to be detected in the current time period.
[0095] For example, the formula corresponding to the prominence of the conduction relationship between any two cylinders to be tested can be:
[0096]
[0097] Among them, G ix is the saliency of the conduction relationship between the i-th cylinder to be tested and the x-th cylinder to be tested. i and x are the serial numbers of different cylinders to be tested. norm() is the normalization function. n is the number of preset dimensions. j is the serial number of the preset dimension. β is a pre-set factor greater than 0, mainly used to prevent the denominator from being 0. For example, β can be 0.001. H ij is the abnormal outstanding index of the i-th cylinder to be tested under the j-th preset dimension. xj Δk is the abnormal prominence index of the x-th cylinder to be tested under the j-th preset dimension. ixj Characterizes the difference in data change between the i-th cylinder to be tested and the x-th cylinder to be tested in the j-th preset dimension. || is the absolute value function. θ ix is the angle between the total force direction corresponding to the i-th cylinder to be tested and the total force direction corresponding to the x-th cylinder to be tested. π is the radian value of 180°. M is the number of acquisition moments in the current time period. b is the sequence number of the acquisition moment in the current time period. k ijb k is the target slope corresponding to the sensor data of the jth preset dimension collected at the bth collection time of the i-th cylinder to be tested in the current time period. xjb It is the target slope corresponding to the sensor data of the x-th cylinder to be detected under the j-th preset dimension collected at the b-th collection time in the current time period.
[0098] It should be noted that, in general, the closer the angle between the force directions of two cylinders on the same cylinder bracket is to 0 or π, the greater the mutual influence between the forces of the two cylinders. When Δk is larger, it often means that the angle between the force direction of the i-th cylinder to be tested and the x-th cylinder to be tested is closer to 0 or π, which often means that the mutual influence of the force between the i-th cylinder to be tested and the x-th cylinder to be tested is greater. ixj The smaller it is, the smaller the data change difference between the i-th cylinder to be tested and the x-th cylinder to be tested in the j-th preset dimension is. ij When H is larger, it often means that the abnormal score of the sensor data under the jth preset dimension obtained by the abnormality detection algorithm can better represent the actual situation of cylinder abnormality. ir The larger the value, the more likely it is that the abnormality score of the sensor data under the rth preset dimension obtained by the abnormality detection algorithm can better represent the actual situation of the cylinder abnormality. ixThe larger it is, the greater the influence between the i-th cylinder to be detected and the x-th cylinder to be detected, and the abnormal scores of the sensor data of the i-th cylinder to be detected and the x-th cylinder to be detected obtained by the abnormality detection algorithm can often better represent the actual situation of the cylinder abnormality.
[0099] Step S4, based on the current synergy coefficient between the cylinders to be detected in the preset dimensions and the prominence of the conduction relationship between the cylinders to be detected, the initial abnormality score corresponding to each sensor data is adaptively corrected to obtain the target abnormality score corresponding to each sensor data.
[0100] As an example, this step may include the following steps:
[0101] In the first step, any sensor data is determined as label data, the cylinder to be detected to which the label data belongs is determined as a reference cylinder, and the preset dimension to which the label data belongs is determined as a reference dimension.
[0102] In the second step, the target abnormality score corresponding to the above-mentioned marked data is determined based on the initial abnormality score corresponding to the above-mentioned marked data, the prominence of the conduction relationship between the above-mentioned reference cylinder and other cylinders to be detected, and the current synergy coefficient between the above-mentioned reference cylinder in the above-mentioned reference dimension and other preset dimensions.
[0103] For example, the formula for determining the target anomaly score corresponding to the sensor data can be:
[0104] E ijb =norm(S ijb ×(w ijb1 +w ijb2 ));
[0105]
[0106] Among them, E ijb is the target anomaly score corresponding to the sensor data of the jth preset dimension collected by the i-th cylinder to be detected at the b-th collection time in the current time period. The preset dimension to which the sensor data belongs is the j-th preset dimension, and the cylinder to be detected to which the sensor data belongs is the i-th cylinder to be detected. i and x are the serial numbers of different cylinders to be detected. b is the serial number of the collection time in the current time period. j and r are the serial numbers of different preset dimensions. norm() is the normalization function. S ijb is the initial abnormal score corresponding to the sensor data of the jth preset dimension collected at the bth collection time of the i-th cylinder to be tested in the current time period. ijb1 and w ijb2 It is used to correct S ijb Two weights; N is the number of cylinders to be tested.xjb G is the initial abnormality score corresponding to the sensor data of the xth cylinder to be tested under the jth preset dimension collected at the bth collection time in the current time period. ix is the salience of the conduction relationship between the i-th cylinder to be tested and the x-th cylinder to be tested. n is the number of preset dimensions. S irb is the initial abnormal score corresponding to the sensor data of the i-th cylinder to be detected under the r-th preset dimension collected at the b-th collection time in the current time period. ijr is the current coordination coefficient between the jth preset dimension and the rth preset dimension of the i-th cylinder to be detected.
[0107] It should be noted that when μ ijr When the value is larger, it often indicates that the correlation between the rth preset dimension and the jth preset dimension is greater, and it often indicates that the calculated abnormal scores of the sensor data under the rth preset dimension and the jth preset dimension can better represent the actual situation of the cylinder abnormality, which often indicates that S irb The more it can represent the actual abnormality of the cylinder to be tested. irb ×μ ijr The larger the irb and μ ijr When both are large, it often means that the sensor data under the preset dimension related to the jth preset dimension can better represent the real abnormal situation, and often means that the cylinder to be detected is more likely to be abnormal, and often means that the cylinder abnormality warning is more needed, and often means that the abnormality score corresponding to the sensor data under the jth preset dimension collected at the bth collection time in the current time period for the i-th cylinder to be detected should be relatively large. ix When the value is larger, it often indicates that the influence between the i-th cylinder to be detected and the x-th cylinder to be detected is greater, and it often indicates that the calculated abnormal scores of the sensor data under the i-th cylinder to be detected and the x-th cylinder to be detected can better represent the actual situation of the cylinder abnormality, which often indicates that S xjb The more it can represent the actual abnormality of the cylinder to be tested. xjb ×G ix The larger the xjb and G ix When both are large, it often means that the sensor data of the cylinder to be detected related to the i-th cylinder to be detected can better represent the real abnormal situation, and often means that the cylinder to be detected related to the i-th cylinder to be detected is more likely to have an abnormality, and often means that the i-th cylinder to be detected is more likely to have an abnormality, and often means that the cylinder abnormality warning is more needed, and often means that the abnormality score corresponding to the sensor data of the j-th preset dimension collected by the i-th cylinder to be detected at the b-th collection time in the current time period should be relatively large. Therefore, E ijbIt can indicate that the sensor data under the jth preset dimension collected by the ith cylinder to be detected at the bth collection moment in the current time period is a real abnormality of the cylinder. The larger its value is, the more likely it is that the sensor data under the jth preset dimension collected by the ith cylinder to be detected at the bth collection moment in the current time period is real cylinder abnormality data rather than noise data, which often indicates that the cylinder to be detected is more likely to have an abnormality.
[0108] Step S5: performing anomaly detection on all sensor data according to the target anomaly scores corresponding to all sensor data.
[0109] As an example, if the target anomaly score corresponding to the sensor data is greater than a preset anomaly threshold, the sensor data may be determined to be abnormal data. The preset anomaly threshold may be a pre-set threshold. For example, the preset anomaly threshold may be 0.7.
[0110] Optionally, if abnormal data exists in the abnormal detection result of abnormal detection of all sensor data, it is determined that there is an abnormal cylinder in the cylinder to be detected installed on the target cylinder bracket, and an early warning can be issued to the staff.
[0111] Optionally, if there is sensor data in the sensor data at the current moment whose target abnormality score is greater than a preset abnormality threshold, it can be determined that there is an abnormal cylinder to be detected on the target cylinder bracket at the current moment, and abnormal reminder information can be sent to the staff to warn the staff.
[0112] In summary, compared with directly performing anomaly detection on sensor data through anomaly detection algorithm, the present invention comprehensively considers the relationship between preset dimensions and the relationship between cylinders to be detected, such as the current synergy coefficient and the prominence of the conduction relationship, and adaptively corrects the initial anomaly score corresponding to each sensor data. This can avoid misjudging noise data as sensor data of cylinder abnormality to a certain extent, thereby improving the accuracy of anomaly detection on sensor data, and further improving the accuracy of anomaly detection on cylinders to be detected.
[0113] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, a person skilled in the art should understand that the technical solutions described in the aforementioned embodiments can still be modified, or some of the technical features can be replaced by equivalents. These modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.
Claims
1. A multi-source sensor data intelligent processing method based on a cylinder bracket, characterized in that: The following steps are involved: Obtain sensor data of each preset dimension collected by each cylinder to be detected installed on the target cylinder bracket within the current time period, and determine the initial anomaly score corresponding to each sensor data through an anomaly detection algorithm; According to the abnormal prominent index of each cylinder to be detected in each two preset dimensions and all sensor data of each cylinder to be detected in each two preset dimensions collected in the current time period, the current coordination coefficient between each cylinder to be detected and each two preset dimensions is determined, wherein the abnormal prominent index is characterized by the stability of the sensor data in the preset dimensions; Determine the prominence of the conduction relationship between each two cylinders to be tested according to the abnormal prominence index of each two cylinders to be tested in all preset dimensions and the angle between the total force directions corresponding to each two cylinders to be tested; According to the current synergy coefficient between the cylinders to be detected in the preset dimensions and the prominence of the conduction relationship between the cylinders to be detected, the initial anomaly score corresponding to each sensor data is adaptively corrected to obtain the target anomaly score corresponding to each sensor data; Anomaly detection is performed on all sensor data according to the target anomaly scores corresponding to all sensor data.
2. According to claim 1, a multi-source sensor data intelligent processing method based on a cylinder support is characterized in that: The method for obtaining the abnormal prominent index of each cylinder to be tested in each preset dimension includes: According to all sensor data in each preset dimension collected by each cylinder to be detected in the current time period, an abnormal prominent index of each cylinder to be detected in each preset dimension is determined.
3. The method for intelligently processing multi-source sensor data based on a cylinder support according to claim 2 is characterized in that: The method of determining the abnormal prominent index of each cylinder to be detected in each preset dimension according to all sensor data in each preset dimension collected by each cylinder to be detected in the current time period includes: The absolute value of the difference between each adjacent two sensor data in each preset dimension collected by each cylinder to be detected in the current time period is determined as the target data difference, so as to form a target data difference set of each cylinder to be detected in each preset dimension; The mean of all target data differences in the target data difference set of each cylinder to be detected under each preset dimension is determined as the data representative difference of each cylinder to be detected under each preset dimension, wherein the data representative difference represents the average fluctuation degree of the sensor data under the preset dimension; The information entropy of all sensor data in each preset dimension collected by each cylinder to be detected in the current time period is determined as a data confusion factor in each preset dimension of each cylinder to be detected, wherein the data confusion factor represents the confusion of the sensor data in the preset dimension; Determine the range of all sensor data in each preset dimension collected by each cylinder to be detected in the current time period as the target range of each cylinder to be detected in each preset dimension, wherein the target range represents the maximum fluctuation degree of the sensor data in the preset dimension; According to the data representative difference, data confusion factor and target range of each cylinder to be tested in each preset dimension, the abnormal prominence index of each cylinder to be tested in each preset dimension is determined, wherein the data representative difference, data confusion factor and target range are all negatively correlated with the abnormal prominence index.
4. The method for intelligently processing multi-source sensor data based on a cylinder support according to claim 1 is characterized in that: The method of determining the current coordination coefficient between each of the cylinders to be detected and each of the preset dimensions according to the abnormal prominent index of each of the cylinders to be detected in each of the preset dimensions and all sensor data of each of the cylinders to be detected in each of the preset dimensions collected in the current time period includes: Determine any one of the cylinders to be detected as a marked cylinder, and respectively determine any two preset dimensions as a first marked dimension and a second marked dimension; Taking a time series consisting of all sensor data under a first marking dimension collected by the marked cylinder in a current time period as a first data series; Taking a time series consisting of all sensor data in a second marking dimension collected by the marked cylinder in a current time period as a second data series; determining the absolute value of the Pearson correlation coefficient between the first data sequence and the second data sequence as a dimensional similarity factor between the first marking dimension and the second marking dimension of the marked cylinder; Based on the dimensional similarity factor between the marked cylinder in the first marked dimension and the second marked dimension, and the abnormal prominence index of the marked cylinder in the first marked dimension and the second marked dimension, the current synergy coefficient between the marked cylinder in the first marked dimension and the second marked dimension is determined, wherein the abnormal prominence index and the dimensional similarity factor are both positively correlated with the current synergy coefficient.
5. The method for intelligently processing multi-source sensor data based on a cylinder support according to claim 1 is characterized in that: The method of determining the prominence of the conduction relationship between each two cylinders to be detected according to the abnormal prominence index of each two cylinders to be detected in all preset dimensions and the angle between the total force directions corresponding to each two cylinders to be detected includes: The acquisition time corresponding to the sensor data is used as the horizontal coordinate, and the sensor data is used as the vertical coordinate to construct the coordinate point corresponding to the sensor data; Performing curve fitting on coordinate points corresponding to all sensor data in each preset dimension collected by each cylinder to be detected in the current time period to obtain a data fitting curve of each cylinder to be detected in each preset dimension; The slope of the coordinate point corresponding to each sensor data on the data fitting curve to which it belongs is determined as the target slope corresponding to each sensor data, wherein the target slope represents the fluctuation change of the sensor data; The prominence of the conduction relationship between each two cylinders to be detected is determined based on the abnormal prominence indicators of each two cylinders to be detected in all preset dimensions, the angle between the total force directions corresponding to each two cylinders to be detected, and the target slope corresponding to all sensor data collected by each two cylinders to be detected in the current time period.
6. The method for intelligently processing multi-source sensor data based on a cylinder support according to claim 5 is characterized in that: The formula corresponding to the prominence of the conduction relationship between any two cylinders to be tested is: Among them, G ix is the salience of the conduction relationship between the i-th cylinder to be tested and the x-th cylinder to be tested; i and x are the serial numbers of different cylinders to be tested; norm() is the normalization function; n is the number of preset dimensions; j is the serial number of the preset dimension; β is a pre-set factor greater than 0; H ij is the abnormal outstanding index of the i-th cylinder to be tested under the j-th preset dimension; H xj is the abnormal prominence index of the xth cylinder to be tested under the jth preset dimension; Δk ixj Characterizes the difference in data change between the i-th cylinder to be tested and the x-th cylinder to be tested in the j-th preset dimension; || is the absolute value function; θ ix is the angle between the total force direction corresponding to the i-th cylinder to be tested and the total force direction corresponding to the x-th cylinder to be tested; π is the radian value of 180°; M is the number of acquisition moments in the current time period; b is the sequence number of the acquisition moment in the current time period; k ijb is the target slope corresponding to the sensor data of the i-th cylinder to be tested under the j-th preset dimension collected at the b-th collection time in the current time period; k xjb It is the target slope corresponding to the sensor data of the x-th cylinder to be detected under the j-th preset dimension collected at the b-th collection time in the current time period.
7. The method for intelligently processing multi-source sensor data based on a cylinder support according to claim 1 is characterized in that: According to the current synergy coefficient between the cylinders to be detected in the preset dimensions and the prominence of the conduction relationship between the cylinders to be detected, the initial abnormality score corresponding to each sensor data is adaptively corrected to obtain the target abnormality score corresponding to each sensor data, including: Determine any sensor data as label data, determine the cylinder to be detected to which the label data belongs as a reference cylinder, and determine the preset dimension to which the label data belongs as a reference dimension; The target abnormality score corresponding to the labeled data is determined based on the initial abnormality score corresponding to the labeled data, the prominence of the conduction relationship between the reference cylinder and other cylinders to be detected, and the current coordination coefficient between the reference cylinder and other preset dimensions.
8. The method for intelligently processing multi-source sensor data based on a cylinder support according to claim 7 is characterized in that: The formula for the target anomaly score corresponding to the sensor data is: Among them, E ijb is the target anomaly score corresponding to the sensor data of the i-th cylinder to be detected under the j-th preset dimension collected at the b-th collection time in the current time period. The preset dimension to which the sensor data belongs is the j-th preset dimension, and the cylinder to be detected to which the sensor data belongs is the i-th cylinder to be detected; i and x are the serial numbers of different cylinders to be detected; b is the serial number of the collection time in the current time period; j and r are the serial numbers of different preset dimensions; norm() is the normalization function; S ijb is the initial abnormal score corresponding to the sensor data of the i-th cylinder to be tested under the j-th preset dimension collected at the b-th collection time in the current time period; w ijb1 and w ijb2 It is used to correct S ijb The two weights of; N is the number of cylinders to be tested; S xjb is the initial abnormality score corresponding to the sensor data of the xth cylinder to be tested under the jth preset dimension collected at the bth collection time in the current time period; G ix is the salience of the conduction relationship between the i-th cylinder to be tested and the x-th cylinder to be tested; n is the number of preset dimensions; S irb is the initial abnormal score corresponding to the sensor data of the i-th cylinder to be detected under the r-th preset dimension collected at the b-th collection time in the current time period; μ ijr is the current coordination coefficient between the jth preset dimension and the rth preset dimension of the i-th cylinder to be detected.
9. The method for intelligently processing multi-source sensor data based on a cylinder support according to claim 1, characterized in that: The method of performing anomaly detection on all sensor data according to the target anomaly scores corresponding to all sensor data includes: If the target anomaly score corresponding to the sensor data is greater than the preset anomaly threshold, the sensor data is determined to be abnormal data.
10. The method for intelligently processing multi-source sensor data based on a cylinder support according to claim 1, characterized in that: The method further comprises: If abnormal data exists in the abnormal detection result of abnormal detection of all sensor data, it is determined that there is an abnormal cylinder among the cylinders to be detected mounted on the target cylinder bracket.
